Alf van der Poorten and Continued Fractions

Alf van der Poorten and Continued Fractions
Jeffrey Shallit
School of Computer Science, University of Waterloo
Waterloo, Ontario N2L 3G1, Canada
[email protected]
http://www.cs.uwaterloo.ca/~shallit
1/1
Why Focus on Continued Fractions?
“It is notorious that it is generally damnably difficult to explicitly
compute the continued fraction of a quantity presented in some
other form...”
— VDP (1992)
I
VDP loved continued fractions
I
They feature in at least 1/4 of all his published papers...
I
I
...including his most-cited papers
He obtained many interesting and novel results about them
2/1
VDP’s Continued Fraction Themes
I
Use the 2 × 2 matrix formulation of continued fractions
I
Work with continued fractions of formal Laurent series; then
“specialize” by setting X = c
I
or reduce mod p
I
Use tools to manipulate expansions; e.g., Raney’s theorem
I
Use the “folding lemma” of Mendès France
3/1
Basics of Continued Fractions
I
a continued fraction is an object of the form
b1
α = a0 +
a1 +
b2
a2 + · · ·
I
A continued fraction is said to be simple if the numerators bi
are all 1. In this case we abbreviate it by [a0 , a1 , a2 , . . .].
I
For real numbers α, the simple continued fraction expansion is
essentially unique if the partial quotients ai are integers ≥ 1
for i ≥ 1;
I
rational numbers have terminating expansions;
irrational numbers have infinite, nonterminating expansions;
4/1
Expansions for quadratic numbers
I
an expansion is ultimately periodic iff it is the root of a
quadratic equation.
I
Examples:
I
√
I
√
I
2 = [1, 2, 2, 2, . . .];
7 = [2, 1, 1, 1, 4, 1, 1, 1, 4, 1, 1, 1, 4, . . .];
√
9+2 39
15
= [1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4, . . .].
5/1
Expansions of some famous numbers
Some familiar numbers have expansions with an obvious pattern:
I
e = [2, 1, 2, 1, 1, 4, 1, 1, 6, 1, 1, 8, . . .];
I
e 1/2 = [1, 1, 1, 1, 5, 1, 1, 9, 1, 1, 13, . . .];
I
tan 1 = [1, 1, 1, 3, 1, 5, 1, 7, 19, 1, 11, . . .];
I
tanh 1 = [0, 1, 3, 5, 7, 9, 11, 13, . . .];
I
e 2 = [7, 2, 1, 1, 3, 18, 5, 1, 1, 6, 30, 8, 1, 1, 9, . . .]
6/1
Expansions of some famous numbers
But for some numbers no one knows an easy-to-describe pattern:
I
π = [3, 7, 15, 1, 292, . . .];
I
√
3
I
γ = [0, 1, 1, 2, 1, 2, 1, 4, 3, 13, . . .];
I
ζ(3) = [1, 4, 1, 18, 1, 1, 1, 4, 1, 9, . . .]
I
log 2 = [0, 1, 2, 3, 1, 6, 3, 1, 1, 2, 1, 1, 1, . . .]
2 = [1, 3, 1, 5, 1, 1, 4, 1, 1, 8, . . .];
7/1
My introduction to VDP: Apéry and ζ(3)
“A proof that Euler missed ... Apéry’s proof of the irrationality of
ζ(3)” Math. Intelligencer 1 (1979), 195–203.
- VDP’s most cited paper (60 citations on MathSciNet; 295 on
Google Scholar)
- a breathless account of Apéry’s June 1978 lecture at the Journées
Arithmétiques announcing a proof the irrationality of ζ(3)
- typical VDP style, e.g.,
“Though there had been earlier rumours of his [Apéry] claiming a
proof, scepticism was general. The lecture tended to strengthen
this view to rank disbelief. Those who listened casually, or who
were afflicted with being non-Francophone, appeared to hear only
a sequence of unlikely assertions.”
8/1
A continued fraction for ζ(3)
In that paper:
6
ζ(3) =
5−
1
117 −
where the general term is
−
64
535 −
729
1436 −
4096
3105 − · · ·
n6
.
34n3 + 51n2 + 27n + 5
9/1
A continued fraction for ζ(2)
Also in that paper:
ζ(2) =
π2
=
6
5
1
3+
16
25 +
81
69 +
135 +
where the general term is
+
256
223 + · · ·
n4
.
11n2 + 11n + 3
Red herring or open problem?
10 / 1
Convergents
I
define p−2 = 0; p−1 = 1; q−2 = 1; q−1 = 0, and
pn = an pn−1 + pn−2
qn = an qn−1 + qn−2
I
for n ≥ 0.
then pn /qn = [a0 , a1 , . . . , an ].
I
the converse is true under suitable hypotheses.
I
therefore
·
pn pn−1
qn qn−1
¸
=
·
pn−1 pn−2
qn−1 qn−2
¸·
an 1
1 0
¸
.
11 / 1
Convergents
I
It follows that
¸
¸
·
¸·
¸ ·
·
an 1
a1 1
a0 1
pn pn−1
···
=
1 0
1 0
1 0
qn qn−1
I
this idea is apparently due to Hurwitz (c. 1917), Frame
(1949), and Kolden (1949), independently — but popularized
by VDP
I
take the transpose to get
¸
¸
·
¸·
¸ ·
·
a0 1
an−1 1
an 1
pn
qn
···
=
1 0
1
0
1 0
pn−1 qn−1
I
and so qn /qn−1 = [an , an−1 , . . . , a1 ], a theorem due to Galois
in 1829.
12 / 1
Convergents
I
take the determinant to get
pn qn−1 − qn pn−1 = (−1)n+1 ,
(1)
a famous identity.
I
divide (1) by qn qn−1 to get
pn pn−1
(−1)n+1
−
=
qn qn−1
qn−1 qn
and hence
¶
X (−1)n+1
X µ pi
pi−1
pn p0
=
−
− ;
=
qn−1 qn
qi
qi−1
qn q0
1≤i≤n
I
Therefore
1≤i≤n
X (−1)n+1
pn
= a0 +
qn
qn−1 qn
1≤i≤n
13 / 1
Convergents
I
now, letting n → ∞, we see that if θ = [a0 , a1 , a2 , . . .] is
irrational, then
X (−1)n+1
θ = a0 +
.
qn−1 qn
i≥1
14 / 1
Continued Fractions for Algebraic Numbers
Theorem. (Bombieri & VDP, 1995) Let γ > 1 be the unique
positive zero of the polynomial f (X ), and suppose
γ = [a0 , a1 , . . .].
Then under some technical conditions we have an+1 = bγn+1 c
where pn /qn denotes the n’th convergent and
¶−1
µ
X
(−1)n+1 f 0 (pn /qn ) qn−1 (−1)n
pn
+ 2
−β
γn+1 =
−
.
qn2
f (pn /qn )
qn
qn
qn
β6=γ: f (β)=0
Example. Let f (X ) = X 3 − 2. Then
√
3
2 = [1, a1 , a2 , . . .]
with
an+1
for n ≥ 0.
¹
qn−1
(−1)n+1 3pn2
−
=
3
3
qn
pn − 2qn
qn
º
15 / 1
Some Unusual Continued Fraction Expansions
2
X
n
2−2 = [1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, . . .]
n≥0
(bounded partial quotients)
X
2−Fn
= [0, 1, 10, 6, 1, 6, 2, 14, 4, 124, 2, 1, 2, 2039,
n≥2
1, 9, 1, 1, 1, 262111, 2, 8, 1, 1, 1, 3, 1, 536870655,
4, 16, 3, 1, 3, 7, 1, 140737488347135, . . .].
(large partial quotients are close to powers of 2)
16 / 1
Some Unusual Continued Fraction Expansions
X
10−n! = [0, 9, 11, 99, 1, 10, 9, 999999999999, 1, 8, 10, 1, 99, 11,
n≥1
72
z
}|
{
9, 99999999999999 · · · 999999999999999999, . . .]
(Liouville’s transcendental number)
X (−1)i
i≥0
bi
= [0, 1, 1, 1, 22 , 32 , 142 , 1292 , 252982 , . . .]
where b0 = 1, and bn+1 = bn2 + bn for n ≥ 0. (Cahen’s constant)
Each of these can be viewed as specializations of continued
fractions for elements of Q((X −1 )): formal Laurent series.
17 / 1
Formal Power Series and Continued Fractions
Theorem. (Artin, 1924)
Let n be an integer. Any formal power series
X
f (X ) =
bi X i ∈ Q((X −1 ))
−∞<i≤n
can be expressed uniquely as a continued fraction
1
f (X ) = a0 +
a1 +
1
a2 + · · ·
= [a0 , a1 , a2 , . . .]
where ai ∈ Q[X ] for i ≥ 0 and deg ai > 0 for i > 0.
18 / 1
An Example
f (X ) :=
µ ¶
X
2k
1
√
=
2−2k
X −2k−1
2
k
X − 1 k≥0
=
3
5
35 −9
1
X + ···
X −1 + X −3 + X −5 + X −7 +
2
8
16
128
=
[0, X , −2X , 2X , −2X , 2X , −2X , . . .]
Here are the first few convergents to the continued fraction for f :
n
an
pn
qn
0
0
0
1
1
X
1
X
2
−2X
−2X
−2X 2 + 1
3
2X
2
−4X + 1
−4X 3 + 3X
4
−2X
3
8X − 4X
8X 4 − 8X 2 + 1
19 / 1
An Example (continued)
It can be proved that
I
pn (X ) = (−1)bn/2c Un−1 (X );
I
qn (X ) = (−1)bn/2c Tn (X );
where T and U are the Chebyshev polynomials of the first and
second kinds, respectively.
20 / 1
The Folding Lemma
Lemma. Let
I
w denote the word a1 , a2 , . . . , an ,
I
let −w denote the word w with all terms negated, and
I
let w R denote the reversed word.
Then
pn
(−1)n
qn−1
+
] = [a0 , w , x, −w R ].
= [a0 , w , x −
2
qn
xqn
qn
21 / 1
Proof of the Folding Lemma
Proof. We have
[a0 , w , x
·
−
qn−1 /qn ] ↔
=
·
↔
pn (−1)n
.
+
qn
xqn2
pn pn−1
qn qn−1
¸·
x − qn−1 /qn 1
1
0
xpn − (pn qn−1 − pn−1 qn )/qn pn
xqn
qn
¸
¸
22 / 1
Application of the folding lemma
Define
hn (X ) = X
X
0≤i≤n
i
X −2 ∈ Q(X −1 ).
We find
h1 (X ) = 1 + X −1 = [1, X ]
h2 (X ) = 1 + X −1 + X −3 = [1, X , −X , −X ]
h3 (X ) = [1, X , −X , −X , −X , X , X , −X ]
h4 (X ) = [1, X , −X , −X , −X , X , X , −X ,
−X , X , −X , −X , X , X , X , −X ].
In general, by the Folding Lemma, if
hi (X ) = [1, Y ],
then
hi+1 (X ) = [1, Y , −X , −Y R ].
23 / 1
Continued fractions and folding paper
So — as explained in VDP’s 2nd most cited paper, entitled
FOLDS! — the terms of this continued fraction are given by
folding a piece of paper repeatedly, then unfolding and reading the
sequence of folds as “up” or “down”.
24 / 1
Symmetry in continued fractions
FOLDS! also contains the following beautiful proof of H. J. S.
Smith of Fermat’s two-square theorem:
Let p = 4k + 1 be a prime. Consider the continued fraction
expansion of p/n for 2 ≤ n ≤ (p − 1)/2:
p
= [a1 , . . . , ar ].
n
Then a1 ≥ 2 and ar ≥ 2. By the result mentioned previously,
p
[ar , . . . , a1 ] = 0
n
for some integer n0 with 2 ≤ n0 ≤ (p − 1)/2.
This gives an involution of the (p − 1)/2 − 1 = 2k − 1 numbers
from 2 to (p − 1)/2; so some integer x is paired with itself.
Then
p
= [a1 , . . . , ar ] = [ar , . . . , a1 ].
x
25 / 1
Symmetry in continued fractions
If it’s a palindrome of odd length, say
[b1 , . . . , bs , b, bs , . . . , b1 ]
then, using the pairing
[b1 , . . . , bs ] ↔
·
a a0
b b0
¸
·
a a0
b b0
¸·
we see
p
x
= [b1 , . . . , bs , c, bs , . . . , b1 ] ↔
¸
·
a2 c + 2aa0
abc + ab 0 + a0 b 0
,
=
abc + ab 0 + a0 b
b 2 c + 2bb 0
b 1
1 0
¸·
a b
a0 b 0
¸
a contradiction, since we cannot have
p = a2 c + 2aa0 = a(ac + 2a0 ).
26 / 1
Symmetry in continued fractions
So the palindrome must be of even length, say,
¸
¸·
·
p
a b
a a0
= [b1 , . . . , bs , bs , . . . , b1 ] ↔
a0 b 0
b b0
x
¸
· 2
a + a0 2 ab + a0 b 0
,
=
ab + a0 b 0 b 2 + b 0 2
expressing p as the sum of two squares.
Taking the determinant, we get p(b 2 + b 0 2 ) − x 2 = 1, or x 2 ≡ −1
(mod p) — so we get a square root of −1 (mod p), for free!
27 / 1
Specialization
The expansion
X
i
X
X −2 = [1, X , −X , −X , −X , X , X , −X , −X , X , −X , −X , . . .
0≤i≤n
= [a0 , a1 , a2 , . . .]
is atypical in several respects:
I
the partial quotients ai have integer coefficients;
I
the coefficients lie in a finite set;
I
the partial quotients (except the first) are all linear monomials
in X .
I
the partial quotients are given by a finite automaton
28 / 1
A MSD-first automaton for the partial quotients
1
0
1
1
0
0
-X
X
1
-X
0
X
0
1
1
29 / 1
Specialization
Now “specialize”, setting X = 2. We get
X
i
2
2−2 = [1, 2, −2, −2, −2, 2, 2, −2, −2, 2, · · · ],
i≥0
which is an “illegal” expansion ... so we need to “remove” the
−2’s somehow.
This can be done with the following
Lemma.
[a, −b, c] = [a − 1, 1, b − 2, 1, c − 1];
[a, 0, b] = [a + b].
30 / 1
Application of the Folding Lemma
Thus
2
X
2−2
i
=
[1, 2, −2, −2, −2, 2, 2, −2, −2, 2, −2, · · · ]
=
[1, 1, 1, 0, 1, −3, −2, 2, 2, −2, −2, 2, −2, · · · ]
i≥0
=
=
=
=
=
=
=
···
[1, 1, 2, −3, −2, 2, 2, −2, −2, 2, −2, · · · ]
[1, 1, 1, 1, 1, 1, −3, 2, 2, −2, −2, 2, −2, · · · ]
[1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 2, −2, −2, 2, −2, · · · ]
[1, 1, 1, 1, 2, 1, 1, 1, 2, −2, −2, 2, −2, · · · ]
[1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 0, 1, −3, 2, −2, · · · ]
[1, 1, 1, 1, 2, 1, 1, 1, 1, 2, −3, 2, −2, · · · ]
[1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, −2, · · · ]
31 / 1
Application of the Folding Lemma
More generally, we have the following
Theorem. (VDP and JOS, 1990)
Let a0 = 1, ai = ±1 for i ≥ 1.
Then the number
2
X
ai 2−2
i
i≥0
is transcendental and its continued fraction expansion consists
solely of 1’s and 2’s.
32 / 1
Liouville’s Transcendental Number
Let fn (X ) =
X
1≤k≤n
X −k! , and define f (X ) = f∞ (X ).
In 1851, Liouville proved that f (b) is transcendental for all integers
b ≥ 2.
We can apply the folding Lemma to this number; we get
f1 (X ) = [0, X ]
f2 (X ) = [0, X , −1, −X ] = [0, X − 1, X + 1]
f3 (X ) = [0, X − 1, X + 1, X 2 , −X − 1, −X + 1]
f4 (X ) = [0, X − 1, X + 1, X 2 , −X − 1, −X + 1, −X 12 ,
..
.
X − 1, X + 1, −X 2 , −X − 1, −X + 1]
33 / 1
Liouville’s Transcendental Number
Hence, we get
f (X ) = [0, X − 1, X + 1, X 2 , −X − 1, −X + 1, −X 12 ,
X − 1, X + 1, −X 2 , −X − 1, −X + 1, −X 72 , . . .]
where the large partial quotients are
n! − 2(n − 1)! = (n − 2)(n − 1)!.
34 / 1
An Unusual Continued Fraction
2−1 + 2−2 + 2−3 + 2−5 + 2−8 + 2−13 + · · · + 2−Fn + · · ·
= [0, 1, 10, 6, 1, 6, 2, 14, 4, 124, 2, 1, 2, 2039,
1, 9, 1, 1, 1, 262111, 2, 8, 1, 1, 1, 3, 1, 536870655,
4, 16, 3, 1, 3, 7, 1, 140737488347135, . . .].
35 / 1
A Fibonacci Power Series
Similarly
X −1 + X −2 + X −3 + X −5 + X −8 + · · · + X −Fn + · · · +
= [0, X − 1, X 2 + 2X + 2, X 3 − X 2 + 2X − 1,
−X 3 + X − 1, −X , −X 4 + X , −X 2 ,
−X 7 + X 2 , −X − 1, X 2 − X + 1, X 11 − X 3 ,
−X 3 − X , −X , X , X 18 − X 5 , −X , X 3 + 1, X ,
−X , −X − 1, −X + 1, −X 29 + X 8 , X − 1, . . .]
What’s going on here?
“We remark that to our surprise, and horror, continued fraction
expansion of formal power series appears to adhere to the cult of
Fibonacci.” – VDP (1998)
36 / 1
Theorem. (VDP & JOS, 1993). Let (Fn ) be the sequence of
Fibonacci numbers defined by F0 = 0, F1 = 1, and
Fn+2 = Fn+1 + Fn for n ≥ 0. Let (Ln ) be the sequence of Lucas
numbers, obeying the same recurrence relation, but with L0 = 2
and L1 = 1.
Define
sn = X −1 + X −2 + X −3 + X −5 + · · · + X −Fn = [0, fn ].
Then for n ≥ 11 we have sn+1
R
R
]
, 0, −fn−3 , X Fn−4 , fn−3
= [0, fn , 0, −fn−4 , −X Ln−4 , fn−4
R
R
R
]
, 0, −fn−3 , X Fn−4 , fn−3
, 0, −fn−4 , −X Ln−4 , fn−4
= [0, gn , X Fn−5 , fn−4
R
R
]
, 0, −fn−3 , X Fn−4 , fn−4
= [0, gn , X Fn−5 − X Ln−4 , fn−4
R
= [0, gn+1 , X Fn−4 , fn−3
].
and
s∞ = X −1 + X −2 + · · · + X −Fn + · · · = lim [0, gh ],
n→∞
where
R
gn = fn−1 , 0, −fn−5 , −X Ln−5 , fn−5
, 0, −fn−4 .
37 / 1
The Fibonacci power series
Corollary. A polynomial p is a partial quotient in the expansion of
the Fibonacci power series if and only if p or −p occurs in the
following list:
X + 1;
X 2 ± X + 1;
X 2 + 2X + 2;
X 3 + 1;
X3 + X;
X 3 − X + 1;
X 3 − X 2 + 2X − 1;
X Fn ;
X Ln+2 ;
X Ln+1 − X Fn
38 / 1
The Fibonacci power series
By specialization we get
Corollary. The large partial quotients
2039, 262111, 536870655, 140737488347135, . . .
in the continued fraction expansion of
2−1 + 2−2 + 2−3 + 2−5 + 2−8 + · · ·
differ by 1 from the numbers
2Lh+1 − 2Fh
for h ≥ 4.
39 / 1
An open question
Let f (X ) =
X
n≥0
an X −n ∈ Q[[X −1 ]], where an ∈ {0, 1}.
Give necessary and sufficient conditions for the continued fraction
of f (X ) to have only polynomials with integer coefficients in its
continued fraction expansion.
Now let’s go back to h(X ) = X
P
i≥0 X
−2i
...
40 / 1
Convergents to h(X )
Here is a table of the first few convergents to h(X ):
n
0
1
2
3
4
5
6
an
1
X
−X
−X
−X
X
X
7
8
−X
−X
9
X
pn (X )
1
X +1
−X 2 − X + 1
X3 + X2 + 1
−X 4 − X 3 − X 2 − 2X + 1
−X 5 − X 4 − X 2 + X + 1
−X 6 − X 5 − X 4 − 2X 3
−X + 1
X7 + X6 + X4 + 1
−X 8 − X 7 − X 6 − 2X 5
−X 4 − 2X 3 − 2X + 1
−X 9 − X 8 − X 6 − X 5
−X 4 − 2X 2 + X + 1
qn (X )
1
X
−X 2 + 1
X3
−X 4 − X 2 + 1
−X 5 + X
−X 6 − X 4 + 1
X7
−X 8 − X 6 − X 4 + 1
−X 9 − X 5 + X
41 / 1
Denominators of the Convergents
n
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
an (X )
1
X
−X
−X
−X
X
X
−X
−X
X
−X
−X
X
X
X
qn (X )
1
X
−X 2 + 1
X3
−X 4 − X 2 + 1
−X 5 + X
−X 6 − X 4 + 1
X7
−X 8 − X 6 − X 4 + 1
−X 9 − X 5 + X
X 10 − X 8 − X 4 − X 2 + 1
−X 11 + X 3
−X 12 + X 10 − X 8 − X 2 + 1
−X 13 − X 9 + X
−X 14 − X 12 − X 8 + 1
42 / 1
Properties of the Denominators of the Convergents
Theorem. (Allouche, Lubiw, Mendès France, VDP, JOS)
AllP
of the coefficients of the denominators of the convergents to
i
X i≥0 X −2 lie in {0, ±1}.
Proof. (Sketch)
I
The low-order terms of q2k +n−1 (X ) (i.e., those of degree
< 2k ) are exactly the same as those of q2k −n−1 (X );
I
The high-order terms of q2k +n−1 (X ) are, up to a change of
k
signs of individual terms, equal to X 2 qn−1 (X );
43 / 1
A Converse: The Continuant Tree
We can obtain a converse to the preceding theorem.
Define an infinite labeled binary tree T with root r and node n
labeled L(n), as follows:
I
L(r ) = 1;
I
L(left(n)) = XL(n) + L(parent(n));
I
L(right(n)) = −XL(n) + L(parent(n)).
The paths in this tree consist of the consecutive denominators of
the convergents to the continued fraction
[1, ±X , ±X , ±X , . . .].
44 / 1
The convergents
Theorem.
If a path in T consists entirely of polynomials with coefficients in
{0, ±1}, then it is the sequence of denominators of convergents to
a formal power series of the form
X
i
X
±X −2 .
i≥0
45 / 1
The convergents for h(X )
Theorem. (A, L, MF, vdP, S)
Let
h(X ) = X
X
X −2
i
i≥0
= [a0 , a1 , a2 , . . .]
and set pn /qn = [a0 , a1 , a2 , . . . , an ]. Then
(a) q2n+1 (X ) = Xqn (X 2 );
(b) q2n (X ) = (−1)n (qn (X 2 ) − qn−1 (X 2 ));
(c) The polynomial q2n+1 (X ) is odd;
(d) The polynomial q2n (X ) is even.
46 / 1
The Coefficient Table is Automatic
Theorem. (Allouche, Lubiw, Mendès France, VDP, JOS)
Define cm,n = [X n ]qm (X ), the coefficient of the X n term in the
polynomial qm (X ). Then the double sequence (table) (cm,n )m,n≥0
is automatic.
Here is what a small portion of this infinite table looks like:
m\n
0
1
2
3
4
5
6
7
8
9
0
1
0
1
0
1
0
1
0
1
0
1
0
−1
0
0
0
1
0
0
0
1
2
0
0
−1
0
−1
0
0
0
0
0
3
0
0
0
1
0
0
0
0
0
0
4
0
0
0
0
−1
0
−1
0
−1
0
5
0
0
0
0
0
−1
0
0
0
−1
6
0
0
0
0
0
0
−1
0
−1
0
7
0
0
0
0
0
0
0
1
0
0
8
0
0
0
0
0
0
0
0
−1
0
9
0
0
0
0
0
0
0
0
0
−1
47 / 1
A LSD-First Automaton for the Coefficients of the
Denominators of the Convergents
[1,1]
[0,1]
1
[0,0]
[0,0]
[1,0]
1
[0,1]
[0,1]
[1,1]
[1,0]
[0,1]
[1,0]
0
[0,0],[1,1]
0
[1,1]
[0,0]
[1,1]
0
[1,0]
[0,0],[0,1],
[1,0],[1,1]
[1,0], [0,1]
-1
[0,1]
[1,1]
[0,0]
-1
[0,0] [1,0]
48 / 1
More General Results
We can also consider the formal power series
X
i
g² (X ) :=
²i X −2
h² (X ) := Xg² (X )
i≥0
where ²i = ±1.
Nearly everything we proved for h(X ) also holds for these series. In
particular
I The partial quotients for the continued fractions for g ² (X )
and h² (X ) lie in a finite set;
I The sign sequence (²i )i≥0 is ultimately periodic iff the
corresponding sequence of partial quotients is 2-automatic;
I The denominators of the convergents to g² (X ) and h² (X )
have all their coefficients in the set {0, ±1};
I The sign sequence (²i )i≥0 is ultimately periodic iff the double
sequence of coefficients of the denominators of the
convergents is 2-automatic.
49 / 1
An Open Question
Let the Fibonacci numbers be defined by F0 = 0, F1 = 1, and
Fn = Fn−1 + Fn−2 for n ≥ 2. VDP and JOS considered the
continued fraction for the formal power series
X
r (X ) =
X −Fi = X −1 + X −2 + X −3 + X −5 + X −8 + · · · .
i≥2
We proved the partial quotients all have integer coefficients.
Numerical experiments suggest that the denominators of the
convergents have all their coefficients in {0, ±1}.
Can this be proved?
50 / 1
VDP’s Hadamard Quotient Theorem
Theorem.
Let KPbe a field of characteristic 0. Suppose
P
n
n
n≥0 cn X in K [[X ]] are the expansions of
n≥0 bn X and
rational functions with cn 6= 0 for all n ≥ n0 . If the P
quotients bn /cn
all belong to a finitely generated ring over Z, then n≥n0 bcnn X n is
the expansion of a rational function.
Proof outline by vdp in 4 papers (1982–84).
Full details written down by Robert Rumely (68 pages!) in
1986–87.
51 / 1
Application of the Hadamard Quotient Theorem
Theorem. (H. W. Lenstra, Jr., and JOS)
Let θ be an irrational real number with simple continued fraction
expansion θ = [a0 , a1 , . . .] and convergents pn /qn for n ≥ 0. Then
the following four conditions are equivalent:
(a) (pn )n≥0 satisfies a linear recurrence with constant coefficients;
(b) (qn )n≥0 satisfies a linear recurrence with constant coefficients;
(c) (an )n≥0 is ultimately periodic;
(d) θ is a quadratic irrational.
Since then, a better proof was found by Andrew Granville that
does not depend on HQT, and generalizations by Bézivin.
52 / 1
Some VDP-isms
“This appears to raise a metaphysical problem until continued
fractions come to the rescue.”
“In this report I sanitise (in the sense of ‘bring some sanity to’) the
arguments of earlier reports...”
(in a paper about Schneider’s continued fraction expansions) “Let
it be clearly said that Schneider’s continued fraction barely
warrants even the present limited effort.”
“This example is likely to have first been noticed by persons
excessively interested in Fibonacci numbers.”
(about refereeing) “Happily, here there is no tradition that it is
wrong to be scathing when that is appropriate.”
53 / 1
Farewell VDP
54 / 1
For Further Reading
1. J.-P. Allouche, A. Lubiw, M. Mendès France, A. J. van der
Poorten, and J. Shallit, Convergents of folded continued
fractions, Acta Arithmetica 77 (1996), 77–96.
2. M. Mendès France, A. J. van der Poorten, and J. Shallit, On
lacunary formal power series and their continued fraction
exapnsion, in Number Theory in Progress, Walter de Gruyter,
1999, pp. 321–326.
3. E. Bombieri and A. J. van der Poorten, Continued fractions of
algebraic numbers, in Computational Algebra and Number
Theory, Kluwer, 1995, pp. 137–152.
4. A. J. van der Poorten and J. Shallit, Folded continued
fractions, J. Number Theory 40 (1992), 237–250.
5. A. J. van der Poorten and J. Shallit, A specialised continued
fraction, Canad. J. Math. 45 (1993), 1067–1079.
55 / 1